JPH07219597A - Pitch converting device - Google Patents

Abstract

PURPOSE:To reduce the degradation of a tone quality even though the pitch of the tone is changed. CONSTITUTION:This device is provided with a voice input means 1, a pitch conversion rate input means 2, a linear prediction analyzing means 3 obtaining a linear prediction coefficient by executing the linear prediction analysis of the inputted voice, an inverse filter 4 calculating a residual signal from an obtained linear prediction coefficient and the inputted voice signal, a residual signal re-sampling means 5 interpolating the residual signal with an interpolation polynomial and performing a re-sampling while adjusting a sampling frequency with the conversion rate of a pitch inputted by the pitch conversion rate input means, a synthesizing filter 6 synthesizing a voice using a re-sampled residual signal and the linear prediction coefficient and a synthesized voice output means 7 outputting a synthesized voice the same sampling frequency as that of the time of inputting.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a voice editing / synthesizing device,
The present invention relates to a pitch conversion device such as a sound quality conversion device for changing the pitch and accent of voice.

[0002]

2. Description of the Related Art Conventionally, this type of pitch conversion apparatus adjusts the pitch of a voice by changing the number of revolutions of a tape recorder or the sampling frequency for D / A conversion.

As described above, even in the above-mentioned conventional apparatus, the sound can be converted to a desired pitch by adjusting the reproduction speed.

[0004]

However, in the above-described conventional pitch conversion device, since the reproduction speed of the voice itself is changed, the spectral structure of the voice is destroyed, and the individuality of the voice is lowered, and even the clarity is lowered. There was a problem.

The present invention solves the above-mentioned conventional problems, and an object of the present invention is to provide a pitch conversion device which causes little deterioration in sound quality even when the pitch of a voice is changed.

[0006]

In order to achieve the above-mentioned object, the present invention provides a speech input means, a pitch conversion rate input means, and a linear prediction for linearly predicting an inputted speech to obtain a linear prediction coefficient. Analyzing means, an inverse filter for calculating a residual signal from the obtained linear prediction coefficient and the input voice signal, and an interpolation polynomial for interpolating the residual signal, and a pitch conversion rate input by the pitch conversion rate input means. The residual signal resampling means for performing resampling while adjusting the sampling frequency, the synthesis filter for synthesizing speech using the resampled residual signal and the linear prediction coefficient, and the synthesized speech at the time of input. And a synthetic voice output means for outputting at the same sampling frequency.

In addition to the above structure, the present invention also comprises means for adjusting the time length of voice.

[0008]

The present invention has the following actions due to the above-mentioned structure. That is, generally, when speech is subjected to linear prediction analysis, it is separated into a linear prediction coefficient having speech spectrum information and a residual signal containing pitch information for determining the pitch of speech. Therefore, the speech signal input by the speech input means is subjected to linear prediction analysis by the linear prediction analysis means, the linear prediction coefficient is calculated, and the residual signal is calculated by the inverse filter using the input speech signal and the calculated linear prediction coefficient. If the residual resampling means reduces the pitch of this residual signal based on the pitch conversion rate input by the pitch conversion rate input means, the sampling frequency is set higher than the actual sampling frequency. However, when increasing the pitch, re-sampling is performed at a sampling frequency lower than the actual sampling frequency, the number of samples is increased without changing the shape of the residual signal, and it is calculated by the synthesis filter and linear prediction analysis means. By synthesizing into speech using the linear prediction coefficient obtained, and outputting at the same sampling frequency as the sampling frequency of the input speech, Voice pitch is converted is output.

That is, when the residual signal having pitch information is resampled and the number of samples is changed, the pitch interval becomes wider or narrower at the original sampling frequency. Therefore, the re-sampled residual signal and the linear prediction coefficient calculated from the input speech are used for synthesis by the synthesis filter, whereby the speech with the pitch converted can be synthesized.

Further, the synthesized voice signal has a short time length when the pitch is raised and a long time length when the pitch is lowered, so that the TDHS system and the PICO system are used.
If the time lengths of the input signal and the output signal are matched by combining means for compressing and expanding the voice on the time axis without changing the pitch, such as the LA method, the pitch can be changed without changing the time length of the voice. it can.

[0011]

【Example】

(Embodiment 1) FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention. In FIG. 1, 1 is a voice input means including an A / D converter and the like, 2 is a pitch conversion rate input means, 3
Is a linear prediction analysis means for linearly predicting the input speech to obtain a linear prediction coefficient, 4 is an inverse filter for calculating a residual signal from the linear prediction coefficient and the input speech signal, and 5 is a pitch conversion rate input means. Residual signal resampling means for resampling the residual signal by adjusting the sampling frequency while referring to the pitch conversion rate input in 2, and 6 is a voice using the sampled residual signal and the linear prediction coefficient. A synthesis filter for synthesizing into a signal, and 7 is a synthesis voice input means including an A / D converter for outputting the synthesized voice at the same sampling frequency as that at the time of input.

Next, the operation of the above embodiment will be described with reference to the waveform chart of FIG. In the above example,
Input the voice signal (Fig. 6 (1)) with the voice input means 1,
The voice is converted into a digital signal, and the pitch conversion rate input means 2 inputs a pitch change rate representing how much the pitch of the voice of the input voice signal is changed.

The speech signal input by the speech input means 1 is subjected to linear prediction analysis by the linear prediction analysis means 3 for each signal in a fixed section (hereinafter referred to as a frame) to obtain a linear prediction coefficient. Then, the inverse filter 4 uses the obtained linear prediction coefficient and the voice signal input by the voice input means 1 to generate a residual signal (see FIG. 6).
(2)) is calculated. The calculated residual signal is resampled by the residual resampling means 5 (FIG. 6).
(3)). The resampling frequency at the time of resampling is lower than the sampling frequency of the voice input by the voice input means 1 when the pitch is increased based on the pitch conversion rate input by the pitch conversion rate input means 2. Is set to a value higher than the sampling frequency of the voice when it is lowered, and the value is adjusted by adjusting the value according to the pitch conversion rate.

Naturally, when the resampling is performed, the number of samples of the frame of the residual signal increases or decreases. However, with the increased or decreased number of samples of the residual signal as a new frame length, the residual signal resampled by the synthesis filter 6 is used. The linear prediction coefficient calculated by the linear prediction analysis unit 3 is used to synthesize the speech signal (FIG. 6 (4)). The synthesized voice signal is outputted by the synthesized voice output means 7 at the same sampling frequency as the sampling frequency of the voice inputted by the voice input means 1, whereby the voice whose pitch is converted is outputted.

As described above, according to the first embodiment,
Re-sampling the residual signal with pitch information and changing the number of samples, the pitch interval becomes wider or narrower at the original sampling frequency. By synthesizing the input voice or the calculated linear prediction coefficient with the synthesizing filter, it is possible to synthesize the voice with the converted pitch.

(Second Embodiment) Next, a second embodiment of the present invention will be described. In the above embodiment, the output voice signal is synthesized with the frame length after re-sampling, so when the voice is raised, the time length of the output voice is short,
If it is lowered, the time length of the output sound will be longer (FIG. 6 (4)). Therefore, in this embodiment, means for adjusting the time length of voice is added to the configuration of the first embodiment.

In the configuration shown in FIGS. 2 and 3, the TDHS
A voice time axis expanding means 8 for converting the time length (speed) of the voice without changing the pitch (height) of the voice, such as the PICOLA method or the like, is provided to match the time length of the input signal and the output signal. (FIG. 6 (5)). Also, FIG. 4 and FIG.
In the configuration shown in FIG. 3, a residual time axis compression means 9 for compressing and expanding the residual signal on the time axis by the same method is provided, and the residual signal is compressed and expanded on the time axis to form the frame length of the residual signal. Is combined with the frame length of the input signal and then the voice is synthesized so that the time length of the output voice becomes the same as the time length of the input voice.

That is, the pitch conversion device shown in FIG.
The synthesized voice whose pitch is converted and whose time length is changed is compressed and expanded on the time axis, and is output after matching the time length of the input voice. The pitch conversion apparatus shown in FIG. 3 compresses and expands an input voice signal on the time axis based on the pitch conversion rate before performing the pitch conversion process, and synthesizes after the pitch conversion. The time length of the voice is set to be the same as the input voice signal.
Further, the pitch conversion apparatus shown in FIG. 4 compresses and expands the resampled residual signal on the time axis, matches the frame length of the residual signal with the frame length of the input speech signal, and then synthesizes them with a synthesis filter. By doing so, the time length of the synthesized voice is kept the same as the time length of the input voice. Then, the pitch conversion apparatus shown in FIG. 5 compresses and expands the residual signal before resampling based on the pitch conversion rate based on the pitch conversion rate, and the resampled residual signal outputs the input speech. The time length of the synthesized voice is set to be the same as the frame length of the signal, and the time length of the synthesized voice is set to be the same as the time length of the input voice.

As described above, the second embodiment has an effect that the pitch (height) of the voice can be converted without changing the time length (speed) of the voice.

[0020]

As is apparent from the above embodiment, the present invention sets the residual signal having pitch information to be lower than the sampling frequency of the voice input when the pitch is increased, and is set higher when the pitch is decreased. By setting and adjusting the value according to the conversion rate of the pitch and re-sampling to change the number of samples, at the original sampling frequency, it is as if the residual signal is expanding or contracting in time. Therefore, the pitch interval can be widened or narrowed, and when the resampled residual signal and the linear prediction coefficient calculated from the input speech are used for synthesis by the synthesis filter, the pitch of the voice is converted. This has the advantage that the synthesized speech can be synthesized.

As is apparent from the second embodiment of the present invention, the synthesized voice signal has a short time length when the pitch is raised and a long time length when the pitch is lowered. Therefore, according to the ratio of the pitch conversion of the synthesized voice signal, the input voice before pitch conversion, the residual signal after resampling or the residual signal before resampling, compression and expansion on the time axis, By adjusting the time lengths of the input signal and the output signal, the pitch can be changed without changing the time length of the voice.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing the configuration of a pitch conversion device according to a first embodiment of the present invention.

FIG. 2 is a schematic block diagram showing a configuration of a pitch conversion device according to a second embodiment of the present invention.

FIG. 3 is a schematic block diagram showing a modification of the second embodiment of the present invention.

FIG. 4 is a schematic block diagram showing a modification of the second embodiment of the present invention.

FIG. 5 is a schematic block diagram showing a modification of the second embodiment of the present invention.

FIG. 6 is a waveform diagram for explaining the operation of the embodiment of the present invention.

[Explanation of symbols]

 1 Voice Input Means 2 Pitch Conversion Rate Input Means 3 Linear Prediction Analysis Means 4 Inverse Filters 5 Residual Signal Resampling Means 6 Synthetic Filters 7 Synthetic Speech Output Means 8 Voice Time Axis Expansion / Expansion Means 9 Residual Time Axis Expansion / Expansion Means

Claims (2)

[Claims] 1. A voice inputting means, a pitch conversion rate inputting means, a linear prediction analyzing means for linearly predicting input speech to obtain a linear prediction coefficient, an obtained linear prediction coefficient and an input speech signal. And an inverse filter for calculating a residual signal from the residual signal, the residual signal is interpolated by an interpolation polynomial, and the residual signal is resampled while adjusting the sampling frequency according to the pitch conversion rate input by the pitch conversion rate input means. Pitch comprising sampling means, synthesis filter for synthesizing speech using the resampled residual signal and linear prediction coefficient, and synthetic speech output means for outputting synthesized speech at the same sampling frequency as at input Converter. 2. The pitch conversion device according to claim 1, further comprising means for adjusting the time length of the voice.